import tensorflow as tf
from tensorflow.keras import layers, models

# Sample data - in practice, load from your dataset
texts = tf.constant([
    "i love this product amazing quality", 
    "terrible service very disappointed",
    "excellent customer support helpful staff", 
    "poor quality broke immediately",
    "fantastic experience highly recommend",
    "worst purchase ever complete waste"
])
labels = tf.constant([1, 0, 1, 0, 1, 0], dtype=tf.float32)

# Configuration
max_tokens = 10000
sequence_length = 20
embedding_dim = 64

# Text preprocessing pipeline
text_vectorizer = layers.TextVectorization(
    max_tokens=max_tokens,
    output_mode="int",
    output_sequence_length=sequence_length,
    standardize="lower_and_strip_punctuation"
)

# Fit the vectorizer on training data
text_vectorizer.adapt(texts.batch(8))

# Model architecture
def create_text_classifier():
    # Input: raw strings
    text_input = layers.Input(shape=(), dtype=tf.string, name='text')

    # Convert to integer sequences
    x = text_vectorizer(text_input)  # Shape: (batch, sequence_length)

    # Embed tokens into dense vectors
    x = layers.Embedding(
        input_dim=max_tokens,
        output_dim=embedding_dim,
        mask_zero=True,  # Handle padding
        name='token_embedding'
    )(x)  # Shape: (batch, sequence_length, embedding_dim)

    # Pool across sequence dimension
    x = layers.GlobalAveragePooling1D()(x)  # Shape: (batch, embedding_dim)

    # Classification head
    x = layers.Dense(128, activation='relu')(x)
    x = layers.Dropout(0.2)(x)
    output = layers.Dense(1, activation='sigmoid', name='prediction')(x)

    return models.Model(text_input, output)

# Build and compile model
model = create_text_classifier()
model.compile(
    optimizer='adam',
    loss='binary_crossentropy',
    metrics=['accuracy', 'auc']
)

# Train the model
history = model.fit(
    texts, labels,
    batch_size=2,
    epochs=10,
    validation_split=0.2,
    verbose=1
)

# Inspect learned embeddings
embedding_layer = model.get_layer('token_embedding')
vocab = text_vectorizer.get_vocabulary()

# Get embeddings for specific tokens
token_index = vocab.index('excellent')
token_embedding = embedding_layer.get_weights()[0][token_index]
print(f"Embedding for 'excellent': {token_embedding[:5]}...")

        __
        __